Long-haul implementation of White Rabbit Ethernet for fiber-optic synchronization of VLBI stations Jeroen Koelemeij 2nd ngvla workshop NRAO Socorro, NM, USA December 9, 2015 :
Research aims at VU University Amsterdam Methods for high-accuracy fiber-optic time and frequency (T&F) distribution Key focus: compatibility with existing data/telecom infrastructure and existing TFT protocols Long-term objective: a terrestrial infrastructure for telecom, T&F distribution, and (strongly improved) positioning See also NRAO special colloquium this Thursday
Fiber-optic time and frequency transfer for radio astronomy/vlbi Past: Stand-alone H-masers, telescope data recorded on tape/harddisks, shipped to correlator Present: Stand-alone H-masers, telescope data transferred to correlator through fiber-optic telecom network Future: Both T&F reference and telescope data transferred through fiber-optic telecom network Lower cost Superior time sync (<< 1 ns) Important to many stakeholders (GPS back-up for electricity grids, telecom networks, electronic financial transactions)
Fiber-optic frequency transfer Optical frequency transfer (1.5 µm / 2 10 14 Hz): Send ultrastable continous-wave (CW) clock laser down a long fiber-optic link (hundreds to thousands of kilometers) Remote end application: lock optical frequency comb laser to CW laser intensity Clock laser from fiber link Telecom wavelengths (1.5 µm) 1 GHz frequency Optical output Advantage: highest stability (ADEV 10-15 @1s) Disadvantage: frequency comb price comparable to active H-maser PD N 1 GHz electrical output (N 1-10)
Fiber-optic frequency transfer RF/Microwave frequency transfer (10 MHz 10 GHz) Modulate link laser (e.g. AM) with reference clock signal Remote end: simply detect modulation with photoreceiver (no other laser needed) 200 THz power 1 GHz frequency Need to deal with optical path length changes
Optical path length variations Acoustic and (soil) temperature variations change optical path length noisy Doppler shift Effects investigated* using 2 300 km fiber link between VU Amsterdam and RU Groningen H-maser Stability limit > 10-14 Appl. Opt. 54, 728 (2015)
Optical path length stabilization Compensation of frequency fluctuations due to length fluctuations*: PLL 1.5 µm clock laser Noise detection +compensation *L.-S. Ma, P. Jungner, J. Ye, J.L. Hall, Opt. Lett. 19, 1777(1994) roundtrip contains 2 noise! Optical fiber (>> 100 km) Partial reflector Clock laser + noise power power Laser frequency in Laser frequency out
A few historical results Model of frequency transfer through compensated optical fiber (NIST) P. A. Williams et al., J. Opt. Soc. Am. B 25, 1284 (2008) Compensated fiber-optic link of 1840 km length (PTB Germany): S. Droste et al., Phys. Rev. Lett. 111, 110801 (2013) MW-modulated optical carrier over 86 km (LPL and Observatoire Paris): O. Lopez et al., Eur. Phys. J. D 48, 35 (2008) 9.15 GHz AM 15 Hz BW T&F transfer through 540 km telecom link with live Internet data traffic (LPL & Obs. Paris): O. Lopez, Opt. Express 20, 23518 (2012)
Fiber-optic time transfer Synchronization: transmission delay must be measured t A B t = t +? Tx Optical fiber Rx d (unknown) 1. Determine round-trip delay using clock A 2. Assume identical delays A-B and B-A and compute OWD = RTD/2 3. Send correction to clock B taking into account one-way delay
Differential delay Key assumption in both compensated frequency and time links: the optical delays A-B and B-A are identical This can be violated by: Path length variations faster than round-trip time Differential delay due to separate physical paths of unequal length and/or sensitivity to environmental changes Chromatic dispersion (CD) Polarization mode dispersion (PMD) (every fiber is a randomly varying birefringent medium: propagation delay depends on polarization state of light) Nonlinear (power-dependent) effects (often negligible)
A few examples Chromatic dispersion 0-17 ps/(nm km), depending on fiber type Wavelengths 20 nm apart: Differential delay 2 ns/100 km timing offset! Probably leads to uncompensated frequency noise at 10-15 level PMD is known to cause timing drift in older legacy fiber: Without polarization scrambler Solutions: - Choose nearby wavelengths ( 1 nm) - Eliminate dispersion by calibration* * N. Sotiropoulos et al., Opt. Express 21, 32643 (2013) With polarization scrambler Graph by LPL and Observatoire Paris; see O. Lopez et al., Eur. Phys. J. D 48, 35 (2008)
Fiber-optic time transfer methods Measuring delays: exchange PPS signal (1 measurement/s) Better: use 1-10 Gb/s data (10 10 measurements/s possible) Different methods: Cross-correlation of input and roundtrip data (VU, TU/e) delay g: original signal (pseudo-random bit sequence, PRBS) f : delayed (round-trip) signal Delay of 75 km fiber: 4 ps uncertainty [Sotiropoulos et al., Opt. Express (2013)] Resolution <100 fs possible (both in optical and electrical domain!)
Fiber-optic time transfer methods Measuring delays: exchange PPS signal (1 measurement/s) Better: use 1-10 Gb/s data (10 10 measurements/s possible) Different methods: Cross-correlation of input and roundtrip data (VU, TU/e) White Rabbit Ethernet (CERN, based on IEEE Precision Time Protocol) Atomic clock Time, frequency, and 1 Gb/s data in one 1 PPS, 10 125 MHz Designed for 1 ns timing over distances <10 km (LHC, CERN) Commercially available
White Rabbit Communication generally over two different lambdas Use Wavelength Division Multiplexing Choose your own wavelength: just swap SFP transceiver Pro: T&F connection can be a cascade of links with different wavelength pairs (using optical-electrical-optical converters) Con: cascade becomes noisy as each O-E-O conversion adds jitter For high stability we prefer long spans with optical amplifiers
Extending the range of WR Link VSL Delft- NIKHEF Amsterdam (2 137 km) 300 km
Extending the range of WR 2 137 km 1 PPS 10 MHz 1490 nm 1490 nm Cs clock UTC(VSL) 1470 nm 1470 nm 1470 nm 1470 nm 1 PPS 10 MHz 1 PPS 10 MHz 1490 nm 1490 nm Delay asymmetry (characterized for each component)
Optical amplifiers Need optical amplifiers both inside C-band (EDFAs) and outside C-band (SOAs) SOA EDFA 1300 1400 1500 1600 nm EDFAs are known to work well for ultrastable fiber-optic T&F transfer (NIST, PTB, LPL/Observatoire de Paris ) SOAs offer a much wider wavelength range. Do they work as well?
Comparison EDFA vs SOA Use Hz laser and spooled fiber (5, 25, 100 km) SOA vs EDFA: small differences, only visible with frequency counters in high-precision mode Rigorous (quantitative) analysis of noise in system SOAs (ASE, current noise) EDFAs (ASE) Counters (response*) 5 km uncompensated *E. Rubiola, Rev. Sci. Instr. 76, 054703 (2005) S. T. Dawkins et al., IEEE Trans. Ultrason. Ferroelectr. Freq. Control 54, 918 (2007) 5 km compensated, SOA 5 km compensated, EDFA 5 km compensated, no amp
Results 2 137 km roundtrip Excess jitter due to meaurement system (RMS jitter WR switch: 9 ps) Origin of offset: TBD Time offset 5 ns (within current uncertainty of ±8 ns due to dispersion) * Comparable with state-of-the-art GPS timing Uncertainty currently improved towards ±0.1 ns (0.1 ns c 3 cm ) In principle, few picoseconds uncertainty is possible (demonstrated using 75 km fiber spools by VU, TU Eindhoven and SURFnet**) Work in progress: Tjeerd Pinkert (VU), Henk Peek, Peter Jansweijer (NIKHEF) **E. Dierikx et al. Transactions on Ultrasonics, Ferroelectrics, and Frequency Control **N. Sotiropoulos, C.M. Okonkwo, R. Nuijts, H. de Waardt, JK Optics Express 21, 32643 (2013)
Timing for VLBI ADEV Slave-Master Required clock stability Current WR Hi-quality LO Improved WR(?) Rogers & Moran, IEEE Trans. Instr. Meas. IM-30, 283 (1981) Requires HW temperature monitoring and active compensation (software) - work in progress* ASTERICS project (EUH2020, 15 M ) Coordinated by ASTRON (M. Garett) Subtask: VLBI timing VU, JIVE, ASTRON, NIKHEF, SURFnet, University of Granada (Spain) Improve WR and demonstrate *G. Gong et al. (LHAASO, 2014) DOI: 10.1109/RTC.2014.7097462
Jitter forecast Microsemi 1000C oscillator Oscilloquartz BVA Wenzel Associates SSB PSD [dbrad 2 /Hz] 5 MHz output rms jitter [fs] in interval [1 Hz, f max ] 10% SNR degradation (120 GHz system freq.) 1% SNR degradation (120 GHz system freq.) Frequency [Hz] Economical solution (if it works): WR equipment < $1k - $3k LO < $10k f max [MHz]
Implementation of WR T&F distribution for radio astronomy in live telecom network
Westerbork Dwingeloo (ASTRON) H- maser WRE node Tx Rx 25 km Time & frequency for VLBI WSRT Existing data link New data link Correlator CAMRAS Clean-up oscillator WRE node Tx Rx
Westerbork Dwingeloo (ASTRON) H- maser WSRT WRE node Tx Rx Existing data link 25 km Time & frequency for VLBI New data link Correlator CAMRAS Clean-up oscillator OSC Tx DWDM node Tx DWDM node Rx OSC Rx WRE node OEO Tx Rx OSC band C-band C-band OSC band 2 45 km Groningen Assen OSC Tx DWDM node Tx DWDM node Rx OSC Rx OEO C-band C-band 2 55 km C-band OSC Rx OSC Tx OEO OEO OSC Tx OSC Rx C-band
Westerbork Dwingeloo (ASTRON) H- maser WSRT WRE node Tx Rx 25 km Time & frequency for VLBI and LOFAR Existing data link New data link Correlator CAMRAS Clean-up oscillator OSC Tx DWDM node Tx DWDM node Rx OSC Rx WRE node OEO Tx Rx OSC band C-band C-band OSC band 2 45 km Groningen Assen Rx2 Tx2 65 km Rx WRE node WRE switch Tx Rx1 Clean-up oscillator Tx0 Rx0 Tx1 DF OSC Tx DWDM node Tx DWDM node Rx OSC Rx OEO LOFAR CN Buinen C-band C-band 2 55 km C-band OSC Rx OSC Tx OEO OEO OSC Tx OSC Rx C-band
Transport through DWDM in NM? Technologically feasible Is there an owner/operator of interstate DWDM? (there has got to be!) Can the concept of T&F distribution/gps-backup be escalated to an issue of national importance? Timing also helps 4G operators to upgrade to 4G LTE-TDD!
Can timing help phase calibration? Reduce time window for fringe search? Facilitate tropospherical phase drift retrieval algorithms?
Contributions by (among others) Henk Peek (Nikhef) Erik Dierikx (VSL) Tjeerd Pinkert (VU) Peter Jansweijer (Nikhef) Paul Boven (JIVE) Rob Smets (SURFnet) Chantal van Tour (TUD/VU) and collaborators at: Kjeld Eikema (VU)
Thanks! Questions: j.c.j.koelemeij@vu.nl